Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75609.18069986637,
-      "required_pages_per_second_upper_estimate": 0.006553612153347897
+      "required_token_positions_per_second": 75171.50745852057,
+      "required_pages_per_second_upper_estimate": 0.006515675746060733
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9451.147587483296,
-      "required_pages_per_second_upper_estimate": 0.0037702040798960014
+      "required_token_positions_per_second": 9396.438432315072,
+      "required_pages_per_second_upper_estimate": 0.0037483797799246336
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18902.29517496659,
-      "required_pages_per_second_upper_estimate": 0.0043727133196954995
+      "required_token_positions_per_second": 18792.876864630143,
+      "required_pages_per_second_upper_estimate": 0.004347401319295654
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28353.442762449886,
-      "required_pages_per_second_upper_estimate": 0.001726575905215726
+      "required_token_positions_per_second": 28189.315296945217,
+      "required_pages_per_second_upper_estimate": 0.001716581403676753
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28353.442762449886,
-      "required_pages_per_second_upper_estimate": 0.0021827265112063727
+      "required_token_positions_per_second": 28189.315296945217,
+      "required_pages_per_second_upper_estimate": 0.00217009152457792
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28353.442762449886,
-      "required_pages_per_second_upper_estimate": 0.352054866240981
+      "required_token_positions_per_second": 28189.315296945217,
+      "required_pages_per_second_upper_estimate": 0.3500169524187047
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.140550233101294,
-    "producer_p95_seconds": 0.37611892700078897,
-    "gpu_update_seconds": 0.9751196788213619,
-    "producer_batches_per_second": 7.11489392749213,
-    "gpu_batches_per_second": 1.025515146211295,
-    "producer_headroom": 6.937873081424184,
+    "producer_batch_seconds": 0.14045389598425118,
+    "producer_p95_seconds": 0.3750779670008342,
+    "gpu_update_seconds": 0.980797146321469,
+    "producer_batches_per_second": 7.119774022588366,
+    "gpu_batches_per_second": 1.0195788229508542,
+    "producer_headroom": 6.983054043808396,
     "sustainable": true,
-    "buffer_seconds": 60.45742008692444,
+    "buffer_seconds": 60.80942307193108,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 507631.4843254223,
-    "upload_bytes_per_second": 6214687.544045871,
+    "generation_bytes_per_second": 516957.8484346793,
+    "upload_bytes_per_second": 6194820.534109761,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.92589948007586,
+    "estimated_full_shards_per_hour": 110.92712011127743,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.109467931000836,
-      "bytes": 5783909,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.8641120300017064,
       "bytes": 2329685,
@@ -562,6 +556,12 @@
       "bytes": 6195752,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.7641328610043274,
+      "bytes": 2288843,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -325,3 +325,4 @@
 {"step": 32500, "loss": 0.6430358495563269, "lr": 0.0003, "grad_norm": 0.13856002775264428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4756103473, "tokens_per_second": 149169.28847649784, "data_wait_s": 1.808099477784708e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 32600, "loss": 0.6330348439514637, "lr": 0.0003, "grad_norm": 0.13050141602012488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4770742301, "tokens_per_second": 149874.27247702537, "data_wait_s": 9.467003110330552e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 32700, "loss": 0.6028679460287094, "lr": 0.0003, "grad_norm": 0.13054031573608635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4785354945, "tokens_per_second": 148592.2659081701, "data_wait_s": 1.6526995750609785e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 32800, "loss": 0.6038375701755285, "lr": 0.0003, "grad_norm": 0.14351599564844275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4800038111, "tokens_per_second": 149519.52894199218, "data_wait_s": 1.7172998923342675e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4133881df98b7b8cb589e9c5711070ba43dcf0537caefd77cb5d0b5e8d98ff40
-size 691675
+oid sha256:f863ddc6c9877c47218a3352b7c3ad7d59528bc2d6c1fa6e4655b40f3d0fe049
+size 693793